Condition monitoring of rolling bearings is essential for ensuring the reliability of mechanical systems operating under severe or insufficient lubrication conditions. This study proposes a fault diagnosis framework that integrates tribological interpretation of wear phenomena, acoustic emission (AE) signal analysis, and machine learning, based on bearing life tests conducted under dry conditions as an accelerated wear environment to capture damage progression within a practical experimental time. Unlike conventional studies relying on artificially introduced defects, this work focuses on AE signals obtained from bearings in which damage initiates and progresses through actual wear processes. Life tests were conducted using deep groove ball bearings under two radial load conditions. The temporal evolution of the coefficient of friction, AE signals, and surface damage was analyzed. Although the coefficient of friction was the most sensitive indicator of wear progression, its direct measurement is impractical for in-service applications. Frequency-domain analysis revealed that AE counts per second and band-specific AE energy exhibit early changes consistent with the evolution of the friction coefficient. Using these physically interpretable AE features, a fully connected neural network was developed to classify bearing conditions into normal, early-stage damage, and damage progression. The proposed model achieved an average classification accuracy of approximately 85%, demonstrating the effectiveness of AE-based machine learning for bearing fault diagnosis under real wear progression conditions rather than artificial defect scenarios.
Nakai et al. (2026) studied this question.